Improvement Of Dynamic Lattice Searching Algorithm And Application For Clusters

Clusters play an important role in the field of chemistry,due to their unique properties,such as electronic structure,magnetism and catalysis.To clearly understand cluster properties,determining their structure is a necessary means.At present,there are still many limitations to determine the cluster structure only by experimental methods,such as experimental environment and cost.Therefore,the method of theoretical prediction came into being.The theoretical prediction method can not only determine the existing structure in the experiment,but also further predict the cluster structure that has not been found in the experiment and may has novel characteristics.At present,there are mainly two types of prediction methods for clusters.One is based on the evolution of a single structure,which is like a drunken man across the mountains on the potential energy surface.Another algorithm is based on population evolution.Both algorithms have been applied to various clusters under the development of their researchers.However,they have shortcomings.The single structure evolution method has a great dependence on the initial structure.It is difficult to across higher barriers when falls into local minima.The latter tends to“precocity”and hard to find some deep and narrow funnel.Therefore,more efficient theoretical prediction methods need to be developed.Dynamic lattice search algorithm(DLS)is an efficient and unbiased global optimization algorithm proposed by Shao research group in 2004.The idea of DLS algorithm is based on the existence of some low-energy vacancies on the surface of the cluster(so called lattice).By moving the higher-energy atoms to these vacancies,the structure with lower energy can be obtained.Subsequently,the DLS algorithm was applied to LJ,Morse,Ag and other clusters,and its efficiency was verified.However,the current DLS algorithm cannot be applied to some specific systems,such as metal oxide clusters.In this paper,we first improved the dynamic lattice search(DLS)algorithm and explore the metal oxide cluster structures.Then,we propose a novel global optimization algorithm and apply it to Lennard-Jones(LJ)clusters.The main contents of this paper are as follows:1.Improved DLS algorithm for the optimization of Al₂O₃ clusterThe DLS algorithm is improved and applied to Al₂O₃ clusters.For Al₂O₃ clusters,due to the influence of charge of Al ion and O ion,the lattice points of Al₂O₃ cluster are fixed.O ions are generated near Al ion as the initial lattice,and vice versa.In the process of lattice movement,there are significant difference of efficiency between the different methods.The mode of Al atom and O atom alternate movement is selected.Finally,the improved DLS algorithm is used to search the structures of(Al₂O₃)_1–20 cluster,and the efficiency and cluster structure of the algorithm are discussed.2.Exploring the influence of metal charges for the structure of metal oxide clusters based on the improved DLS algorithmStudying the growth law of clusters is an important means to understand their properties.Coulomb contribution plays a crucial role in the forces inside metal oxide clusters.In order to understand the effect of metal charge with the structure of metal oxide clusters,we combine the improved DLS with the Woodley potential energy function to search the structure of different types of metal oxide clusters by only adjusting the metal charge.In this paper,six kinds of metal oxide clusters under different charges were systematically studied,including M⁺²O?M₃^+2.56O₄?M₂⁺³O₃?M⁺⁴O₂?M₂⁺⁵O₅?M⁺⁶O₃.The total number of atoms range from 2 to 120.The results show that they are more likely to form structures with hanging bonds(M=O)for the clusters with large metal charges,while there is no such fragment in low metal charge systems.3.Increasing Source and Throttling:An interesting Adaptive Waterfall Searching AlgorithmIt has still of widespread concern that how to exploring effectively potential energy surface.To better solving it,the waterfall search(WS)algorithm has developed.The idea of WS method is shunting operation that after the structural disturbance,which can make it explore fully potential energy surface.However,unlimited shunting operation will inevitably lead to a large amount of calculation consumption.It is higher probability that some similar configurations will be obtained by single structure perturbation.So,we add structural similarity checking to suppress the computational consumption caused by similar configurations,that is,'Throttling'.In addition,the success of single structure evolution always depends on the configuration of initial structure.In order to reduce this impact,we set the WS algorithm to multi-structural evolution,namely'Increasing Source'.At this point,WS algorithm can adaptively adjust the search area in the searching process,and can start to evolve from multiple structures.For this WS algorithm with the characteristics of“Increasing Source and Throttling”,we called it as Adaptive Waterfall Search algorithm(AWS).The structural disturbance of WS method draw on the idea of move surface atom in DLS.Yet,it shows more searching capability than DLS.Then,we further studied the performance of AWS algorithms.The results show that the AWS algorithm not only maintains the exploration ability of WS,but also greatly reduces the computational consumption.Subsequently,we use the AWS algorithm to search some clusters of LJ clusters,and compare it with other algorithms.It is obviously found that the AWS algorithm can still find the global minimum with less computational cost,especially in complex systems.